A Study Of Deep Learning Based LoRa Signal Recognition Under Low SNR

In recent years,IoT(Internet of Things)applications and its market demand are increasing rapidly.Traditional Io T wireless communication technologies like Wi-Fi,Bluetooth and Zig Bee have been unable to meet the requirements of large-scale applications deployment.The emergence of LPWAN(Low-power Wide-Area Network)technology makes it possible to implement large-scale and long-distance applications.As the most mature communication technology in LPWAN,LoRa successfully attracts the attention of academia and industry due to its low power consumption and long distance,and it has been widely used in smart cities,smart farms,logistics tracking and so on.Due to the large coverage of the LoRa network and the long communication distance of the nodes,the transmission environment of LoRa signal is more complicated and changeable.When the noise in the environment seriously interferes with the signal,it is difficult for the gateway to correctly detect and demodulate the LoRa signal in the channel,resulting in bad communication quality.Although LoRa can increase the spreading factor of the sending node to achieve reliable data transmission through the Adaptive Date Rate(ADR)method,the increasement in the spreading factor will increase the transmission delay and node energy consumption,and the ADR is not applicable in the changeable network environment.Therefore,in view of the difficulty in recognizing the LoRa signal at the gateway under low signal-to-noise ratio(SNR)environment,resulting in poor network communication quality,this thesis studies the LoRa signal recognition problem in two scenarios of low-latency network and delay-tolerant network.The contents of this thesis include:(1)LoRa signal recognition in low-latency networks.In response to the challenge that the LoRa gateway is difficult to detect LoRa signals under low SNR conditions,this thesis proposes a LoRa signal detection method based on signal cross-correlation.The LoRa signal can be detected accurately when the SNR is-20 d B.On this basis,a low SNR LoRa signal recognition model LT-CNN based on deep learning is designed for the needs of low latency.Compared with the method of increasing the spreading factor in LoRa,the node energy consumption of LT-CNN in the-20 d B SNR environment is reduced by about 30%,and the transmission delay is reduced by about 7.8s,and the packet reception rate can reach 70%,which is 2.8× higher than the existing method.(2)LoRa signal recognition in delay-tolerant networks: For some applications that are delaytolerant but require high recognition rates,this thesis uses a filtering method based on fractional Fourier transform to preprocess low SNR LoRa signals,and design a highprecision low SNR LoRa signal recognition model Fr-CNN.Compared with the method of increasing the spreading factor in LoRa,Fr-CNN reduces the node energy consumption by about 40% in the-20 d B SNR environment,and the packet reception rate can reach 90%,which is 3.6× higher than the existing method.